Introduction
Four open protocols that let software run unattended — proving who it is, discovering what it can call, and paying for what it uses.
Interagentic is four protocols. They are designed to be adopted separately, and nothing here requires the others.
| Protocol | Solves | Status |
|---|---|---|
| Identity | An agent proving who it is to a server that has never seen it | v0.1 |
| Services | A server describing what it can do, in a form an agent can call | v1 |
| Credits | Paying for work, with a human setting the limit | v0.1 |
| Secure calls | Calling an API without the agent ever holding the credential | v0.1 |
The problem they share
Every one of these is solved already — for humans. A person signs up, clicks through OAuth, pastes a key into a dashboard, and enters a card. None of those steps work when the thing making the request is a process that started at 3am with no one watching.
The usual workaround is to give the process your credentials. That fails in three ways at once: the process can now do anything you can, an attacker who reaches its context gets the same, and you cannot tell afterwards which of your agents did what.
What replaces it
An agent gets its own identity, backed by a keypair it generated itself. It proves that identity with a short-lived signed token that any server can verify by fetching a public key. When it needs something that genuinely requires a person — a connected account, a spending limit — it receives a link to hand over, rather than an error.
npx interagentic initNothing in this documentation requires an account with us. The default broker at
id.interagentic.dev is a convenience; the protocol works the same against one
you run yourself.
Where to start
- Building an agent? Start with the quickstart.
- Have an API you want agents to call? Start with publishing a service.
- Want to read the specs? Start with Identity — the other three assume it.